In a recent LinkedIn post, Charlie Hills 🦩 describes a seven-month effort to build a Claude Code system for managing his LinkedIn content. This article reports on Hills’s account of that process; it was written by Enterprise Media, not by Hills. He says the workflow has generated more than 15.2 million impressions since April, while keeping him in charge of selecting ideas and publishing posts.
“Claude can suggest. The call is always mine.”
Human decisions at the center
According to Charlie Hills 🦩, the system gathers ideas from his own thoughts, the performance of earlier posts and emerging trends. A ranking tool then evaluates ideas against his data, while a validator screens out weaker candidates. Hills emphasizes that these tools inform his choices rather than make them for him.
That distinction is central to his description: the workflow automates research and drafting tasks, but he retains editorial control. His post presents human selection as a safeguard against treating automated recommendations as final decisions.
A process built around his voice
Hills says Claude Code reads six folders containing his posts, voice guidance, rules and previous corrections before it starts generating content. He then makes the graphic before drafting the caption, using tools to produce several design options and check for what he calls “AI design slop.”
“So it never starts from a blank prompt.”
As Charlie Hills 🦩 explains, a post-writer drafts captions in his voice, and a separate check assesses whether the result still sounds like him. The sequence is intended to apply lessons from his existing work rather than ask a model to imitate him without context.
Review, publishing and feedback
Before publication, Hills says a final check considers AI disclosure, whether anything reads as AI-generated, and whether the work is his own. He also says he schedules posts in Notion but opens LinkedIn and publishes them himself. Those steps preserve a final human review and publication decision in the workflow he describes.
“After every post, linkedin-eval reads my stats. Then memory/ saves each fix, so I correct it once.”
Hills portrays performance review as a continuing feedback loop: the system reads post statistics and stores corrections for future use. His post does not provide details on how impressions are measured or attribute that reach solely to the software, but it lays out the process he says supports his content operation.
A packaged operating system
Hills says he has spent seven months building the system and that components marked “AI OS” are available to run in Claude Code or Codex. He also promotes the package in the post and notes a November 8 closing date for access. The account offers a practical example of how one creator combines automation with personal review, though its reported reach remains Hills’s own claim.
📝 About This Content
This article is based on insights shared by Charlie Hills 🦩 on LinkedIn.
📅 Originally posted on November 6, 2026 | View original post on LinkedIn →